construction of auxiliary basis for RI-J calculations

[jmb]

Hello,

I am doing many calculations with an ECP / basis set combination which came without an auxiliary basis for use with RI-J approximation.

The calculations work well, and I would like to move to larger systems, thus I need the large speed advantage this provides.

How does one go about constructing an auxbasis?  Is starting with the normal basis and optimizing it to minimize the total energy a sensible procedure?  How do I decide how many basis vectors to include?  Should I optimize both the exponents and contractions simultaneously, or one at a time?  Will varying the auxbasis to minimize the total energy, ensure that the calculations give the result closest to that obtained without the RI-J approximation?

I am attempting to use 'relax' with the following settings in 'control':  (If I am going about this wrongly please advise)

In 'control'  I set:

$basis    file=basis
$jbas     file=auxbasis

$optimize
   internal   off
   cartesian  off
   global     off
   basis      on   logarithm

In the file 'auxbasis' I set:
....
$jbas
*
#...
*
   3  s expopt          contopt
  7.7447900000     -0.26903500000
  6.7188320000      0.60715300000
  2.7444700000      -1.2912950000
   1  s expopt          contopt
 0.65511600000       1.0000000000

Many thanks in advance for sharing your advice and experience?

[christof.haettig]

The release version of TM does not include the functionalities for optimizing jbas auxiliary basis sets.
Try to contact one of the authors of the available jbas sets....
Christof

[jmb]

Many thanks as always for the useful information!  I'm grateful for your help.

Is there any possibility to obtain the unreleased code which does this optimization?  Or do most people write their own manual optimization loops?

[christof.haettig]

I would say, most people use the basis sets available in the libraries...

I don't know which atom/ECP/basis set combination you are using, but maybe you find in the basis set library and jbas
auxiliary basis set optimized for a similar (or maybe only slightly larger) basis. Then you can try to use this. Do a few small
test molecules calculations ridft and dscf (i.e. with and without RI approximation) and check wether the changes in energies
(or better: changes in reaction energies) are acceptable for your purposes. If this is the case, you can use the auxiliary basis
also for larger systems.
Christof